1. Field of the Invention
This invention relates generally to micropattern generation and replication techniques, and, more particularly, to a novel alignment process for accurately aligning a given object with a focused beam or for accurately aligning two given objects with each other, by using serial detection of repetitively patterned alignment marks and special information processing techniques.
2. Description of the Prior Art
In the fabrication of semiconductor integrated circuits and devices, an important goal has been the increased microminiaturization of these circuits and devices in order to produce structures having minimized size and maximized component density, speed, and complexity. At various points during such fabrication processes, it is frequently necessary to align a semiconductor wafer or substrate with a mask or a source of radiation, or both, for processes such as etching, forming ion-implanted regions, or depositing metal patterns. In order to achieve increased microminiaturization of certain types of integrated circuits, an alignment process is needed which will provide very accurate alignment that is compatible with high resolution of pattern dimensions.
This accurate alignment becomes extremely important when it is necessary to overlay patterns, i.e., to align a given sample or target with several patterns in succession. For example, a representative process for forming an integrated circuit might require 10 levels of masking and registration. Since alignment errors become additive in such a succession of alignment procedures, the necessity for highly accurate alignment is manifest. In addition, the requirement for high resolution imposed by submicrometer design rules further emphasizes the criticality of achieving highly accurate level-to-level registration.
There are two types of precision alignment processes which are of particular current interest in the fabrication of microminiaturized circuits, i.e., circuits having structures with one or more dimensions of one micrometer or less. The first such process uses a focused or shaped beam of radiation, such as a laser, electron, or ion beam, and scans the beam across an alignment mark. The back-scattered or secondary radiation that is generated when the primary beam strikes the alignment mark is collected with a suitable detector. The data from the detector is processed and analyzed in order to determine the position of the beam with respect to the alignment mark. Alignment is achieved either by visually matching an alignment signal with a reference signal on a display screen or by using the risetime of the signal produced by the detection means, such as a photocell or video display system, to implement a servo control system.
A system of the type just described is set forth in U.S. Pat. No. 4,056,730, to D. E. Davis et al. In the process of Davis et al, a square shaped beam of electrons is scanned over a registration mark on the surface of a wafer and the back-scattered electrons are detected by a diode detector. The detector then produces a signal which shows a peak when the electron beam passes over each edge of the registration mark. This signal is then processed and analyzed to provide an indication of the location of the registration marks with respect to the electron beam. One disadvantage of a process such as that of Davis et al is that the resolution, or smallest dimension that the process is capable of achieving, is limited by both the diameter and other variable characteristics of the focused beam. If the same electron beam that is subsequently used for the alignment process was also used to fabricate the alignment marks, the mark so formed may have an imperfect or fuzzy edge due to variations in beam current and skewing of the current density profile. Thus, a sharp edge on the alignment mark is not presented for alignment purposes. In addition, these beam imperfections cause further ambiguity when the beam is scanned across the alignment mark and thus interfere with accurate alignment. The ability to tolerate such beam imperfections and alignment mark imperfections is low in a process in which it is typically required that the beam be aligned to a dimension which is a fraction of the beam diameter. For example, in a process in which a target is to be aligned with a pattern on a mask for the purpose of transferring the pattern to the target, it is often necessary to align the beam to within one-tenth of the dimension of a line on the mask.
More specifically, the alignment mark used in such a prior art focused beam technique typically consists of simple patterns, for example, a cross or L-shape, and often has dimensions which are many times larger than the size of the focused beam. However, in a process which detects primarily or only the edge of the alignment mark, the width of the edge is small with respect to the size of the beam and it is therefore difficult to locate the edge. The ideal conditions for such a process would be to have a zero width-edge and an infinitely small beam size so that when the beam is scanned across the edge, the detected signal would show a sharp rise. However, it is not possible to form an edge with a zero thickness. The edge has a finite width which is determined by the process used to form it, and this width is usually related to the beam diameter, for example, being one-half the beam diameter. Thus, it is difficult to very accurately detect the edge of the alignment mark with a focused beam by such prior art processes. In addition, the actual location of the edge of the alignment mark on the target has a degree of uncertainty because of the uncertainties inherent in the processes used to fabricate these marks (e.g. forming a mask and etching). Further, in these prior art processes, there are inherent alignment mark irregularities, which produce noise in the detection signal.
Another disadvantage of such a prior art edgedetection alignment process is that the signal produced from the information collected by the detecting means is not a square waveform with an edge corresponding to an edge of the alignment mark, but rather is a waveform which shows a gradual rise and fall. Thus, determination of the exact position of the alignment mark edge is not possible from such a gradually sloped signal. In addition, the signal generated from the information from the detection means will have variations due to nonuniformity in the beam, such as fluctuation in current and skewing of the current density profile. Furthermore, there are problems associated with the electrical noise present in the detector means, which give rise to the necessity of increasing the signal-to-noise ratio in order to enhance the signal. Excessive low-pass filtering of the signal to remove noise degrades the risetime and results in no net increase in accuracy.
One approach to increasing the signal-to-noise ratio of the detected signal in an edge-detection alignment process is set forth in U.S. Pat. No. 4,123,661, to Edward D. Wolf and Walter E. Perkins, Jr., assigned to the present assignee, in which high atomic number metals or compounds are used as electron beam registration alignment marks on low atomic number substrates in order to enhance the secondary and back-scattered electron video signals. In addition, it may be noted that in the process of Wolf et al, the enhanced signal contrast so produced is further augmented by placing two registration marks sufficiently close together (e.g., at a distance of less than 3 micrometers) that the separation therebetween will provide sharply definable maximum dip and rise slopes in the video signals, which reflect the position of the edges of the alignment marks. While the process of Wolf et al represents a significant improvement over the prior art, it still relies on edge detection for alignment purposes.
Thus, in an edge-detection system, determination of the exact edge of the alignment mark to an accuracy that is a small fraction of the risetime of the alignment signal is difficult and is further complicated by the presence of noise. Finally, in the case where a scanning focused ion beam is used, the ion beam causes sputtering or removal of material from the edge of the alignment mark as it scans over it, which leads to a degradation in edge definition of the alignment mark and consequent deterioration in alignment accuracy.
The second type of precision alignment process of particular current interest in the fabrication of microminiaturized circuits and devices uses a matched pair of patterned marks on two separate objects, for example, a mask and a target. A flood of radiation is passed through the first set of alignment marks toward the second set of alignment marks. When the radiation strikes the second set of alignment marks, it causes either reflected radiation or a secondary emission of radiation from the marks, which is then detected by a suitable detector means. The amount of radiation striking the second set of alignment marks is proportional to the extent of alignment of the first and second sets of alignment marks. When the two sets of alignment marks are perfectly aligned, the maximum amount of radiation will strike the second set of marks. The information from the detector means is processed to produce a signal whose magnitude is related to the amount of alignment of the first and second alignment mark patterns. Such a system using x-rays is set forth in U.S. Pat. No. 3,984,680 to Henry I. Smith.
One disadvantage of such a patterned mark alignment process is that the maximum resolution achievable is limited by the resolution of an element in the pattern of the alignment mark. In addition, there are frequently imperfections in both sets of alignment mark patterns which are inherent in the process steps by which these patterns are formed (e.g., forming a resist pattern, and etching). Thus, when alignment of these two imperfect patterns is attempted, the waveform of the signal produced does not have a perfect maximum, but rather is a "flat" maximum (i.e., a signal with a truncated or flat top). This flat maximum leads to ambiguity in determining the maximized reflected signal and thus the optimized alignment between the two sets of patterned alignment marks. For example, if the two sets of alignment mark patterns are visually aligned, the patterns may appear fuzzy and the exact edge of the patterns cannot be determined.
In a related alignment technique which uses matched sets of patterned alignment marks in the form of diffraction gratings, both sets of alignment marks are illuminated with radiation, such as a laser beam, as described, for example, by D. C. Flanders, H. I. Smith, and S. Austin in the publication entitled "A New Interferometric Alignment Technique", in Applied Physics Letters, Vol. 31, No. 7, Oct. 1, 1977, pp. 426-428. The radiation diffracted from each set of alignment marks is detected, corresponding detection signals are generated, and an optical interference pattern is generated from the detection signals as a means of comparing the position of the two sets of alignment marks. An error signal is then generated which indicates the extent of alignment of the two sets of alignment marks with respect to each other. However, there are potential systematic errors affecting the accuracy of the alignment process, such as the partial dependence of the magnitude of the error signal on the spacing between the grating marks and the possible non-symmetry in the profile of the grooves of the grating as formed. In addition, the diffraction gratings must be formed to have a frequency which is reasonably close to that of the laser light and the spatial frequency of the marks must be uniform or else multiple beams of differing intensities are formed rather than a collimated, sharply defined beam as is desired for alignment purposes. Further, the diffraction grating marks must have optical contrast in order to be effective and must have a structure with a periodicity that is accurately controlled to a fraction of a wavelength of the radiation used. In some cases, these optical and structural requirements may not necessarily be compatible with semiconductor processing techniques which would be necessary to fabricate a semiconductor device. In addition, rotational misalignment of the two diffraction grating marks causes imperfect meshing of the interference pattern and leads to an error signal that has a soft maximum. Further, the separation between the two diffraction gratings must be accurately controlled or else spurious results are obtained. Finally, such a diffraction grating system is mechanically complex, requiring, for example, special mirrors and special alignment procedures therefor, which give rise to systematic errors.
It is the alleviation of many of the above-mentioned problems in prior art alignment processes to which the present invention is directed. More particularly, some of the problems that the present invention seeks to overcome are: the limitations on maximum achievable resolution imposed by process parameters such as beam diameter or resolution of an element of the alignment mark pattern; the inability to accurately detect the edge of an alignment mark; and the inability to generate an accurate alignment signal with a high signal-to-noise ratio.